Poll Dancing

In a working paper earlier this year, Justin Wolfers and I suggested that there was more than sampling error at work in Australian opinion polls, and that they suffered from systemic biases. The latest ACNielsen poll – showing Labor 54% and the Coalition 46% – appears to support the point. I can’t find a sample size reported anywhere, but if we assume 1500, then the poll implies that if an election were held today, Beazley would be a 99.9% chance of becoming PM (for an example of how to convert polls into probabilities, see this table). If you don’t believe Labor’s true odds could possibly be 99.9%, then you have to believe that the ACN sample isn’t a representative slice of the population.

For poll-lovers, Bryan Palmer is tracking the various polls here. For poll-sceptics, Centrebet has a market open for the 2007 election, with the Coalition a 63% chance, and Labor a 37% chance. These probabilities sound more like the real world to me.

This entry was posted in Australian issues. Bookmark the permalink.

8 Responses to Poll Dancing

  1. Guy says:

    As always, it seems Centrebet are closer to the mark. Perhaps the mainstream media should start thinking about publishing Centrebet odds every week instead of opinion polls from the likes of ACNielsen et al.

  2. Andrew Leigh says:

    I agree. It’s strange – the media seemed much more inclined to publish betting odds during the campaign than they do now. Yet we know that the performance gap between markets and polls is greatest early in the election cycle.

  3. Hughie says:

    Gently, boys. The AC Nielsen Poll is based on information that’s focussed on an election held “today”, the Centrebet data is focussed on an event to happen in two years’ time. They’re predicting different things.

    There may or may not be systemic baises in the Polls – I’ll read your paper in detail, Andrew – but this is not evidence of any of it. My bet is that Centrebet are factoring things like the Caolition’s ability to win elections once the real campaign is called, which seems undoubted. They would have to depend on the poll data, among other things …

    If Centrebet was running odds on an election to be held in a week’s time, I’d bet the odds would be more in the ALP’s favour (unless the Poll was Morgan’s, of course :-).

  4. Andrew Leigh says:

    Hughie, do you really believe that Labor’s chance of winning an election held today is better than 999 in 1000?

    Even close to polling day 2004, polls with these sorts of non-credible numbers were being produced. The extreme results don’t only occur a long way out from the poll.

  5. Sachmo says:

    Isn’t it ridiculous when sample sizes and margins of error aren’t published in opinion polls? Any serious political journalist should insist that they’re published.

    This really struck me in the last federal election – I don’t recall reading about the margin of error for any opinion poll in that election.

  6. Andrew Leigh says:

    Sachmo, I couldn’t agree more. In the US, even CNN publishes margins of error.

  7. Sachmo says:

    In a way it’s funny the way newspoll and morgan publish their poll findings and say things like “the ALP is up 1% and the Coalition is down 1%” – when the changes in support or the difference between the two parties is much smaller than the margin of error.

    Perhaps serious political & economic commentators/academics should just refuse to refer to these poll findings unless the margin of error is also given. You never know, it might force a change.

  8. Andrew Leigh says:

    Sachmo, I argued just this in a SMH oped prior to the last election (http://econrsss.anu.edu.au/~aleigh/opinion_fulltext.htm#PollErrors). Then, during the campaign, the point was ignored by just about every journalist – including at the SMH itself.

Comments are closed.